At echosonic, we are revolutionizing microphones by bringing time-series signal processing inside MEMS sensors by using the mechanical element as the computational source. This enables Machine Learning capabilities into the microphone to reduce our dependency on the cloud and other centralized infrastructures for audio recognition. Oh, just a tiny little breakthrough, no big deal.
Original audio never leaves the device
Reduce >85% of the computational load related to audio processing
On-demand and secure data transfer when necessary
Benefits in IoT
Audio-based predictive maintenance
+ Minimal data required
+ Working in noisy environment
+ Untraceable computational demand and power consumption
+ Easy and low-coast implementation
Detect abnormal sound produce by machine
Send real-time warning and/or stop the production
Allow detection and early identification of problems
Recoverable original audio data for later analysis
We are a group of young and energetic physicists, computer scientists, mechanical engineers, and data scientists aiming to disrupt the edge AI market. With robust and continuous university collaborations and public-private cooperation, we are bringing the cutting-edge research of deep technology to human-centric machine learning and building a public forum for disseminating computational technologies.
Backed by the university research, echosonic exploits the dynamic mechanisms of microphones to filter out the important part of audio signals for learning-based processing. The hardware/software audio processing pipeline can be deployed to edge sensors, eliminating privacy and security concern, and draws only 10% of power consumption compared to the legacy devices.
Low power consumption
Exceptional processing speed
Reduce >90% power consumption of intelligent audio processing compared to legacy devices
Privacy and security
Eliminates privacy and security concerns by processing audio signals internally on microphones
Exceptional processing speed by reducing latency in cloud-microphone data transferring